rye.ai

Rye supervision workbench

Wrap agents

Launch Claude Code, Codex CLI, and other coding assistants through Rye so traffic, policy, and audit events are connected.

route
/docs/install/wrap-agents
control point
external supervisor
signals
proxy / policy / audit
01Claude / Codex
02Rye local proxy
03policy + audit
04model API

Why wrapping matters

AI coding agents read environment variables, load local configuration, open files, run commands, and call model APIs. Wrapping starts the agent under Rye supervision from process launch time, when proxy variables and trust settings matter most.

Wrap common CLIs

Start Claude Code through Rye:

exec:bash
rye wrap claude

Start Codex CLI through Rye:

exec:bash
rye wrap codex

You can also install app shims for repeated use:

exec:bash
rye app wrap claude
exec:bash
rye app wrap codex

After a shim is installed, the normal command starts under Rye supervision from the developer's usual shell.

Verify traffic

Use:

exec:bash
rye history --last 1h
exec:bash
rye status

You should see recent model traffic, process attribution, device identity, and policy decisions.

What wrapping does not do

Wrapping does not replace the agent's own permission model. Claude Code, Codex, Cursor, and other tools may still ask for local approvals. Rye sits outside those controls and gives the organization a separate enforcement and audit path.